The state of agentic commerce in mid-2026: 38% discover, 17% buy
Forrester's mid-2026 read on the state of agentic commerce finds 38% of consumers use answer engines to discover products but only 17% buy through them. The gap is trust and machine legibility, and it decides which brands agents select.
TL;DR
Forrester's mid-2026 read on agentic commerce found a 21-point gap: 38% of consumers familiar with answer engines use them to discover products, but only 17% buy through them. The blocker is not interest, it is trust and machine legibility. Answer-engine traffic converts very well, but it only reaches brands an agent can actually parse, which is why Forrester tells merchants to build a machine advantage.
The number that matters: 38% discover, 17% buy
Forrester's mid-2026 read on agentic commerce leads with a gap, not a boom. Among consumers familiar with answer engines, 38% use them to discover products. Only 17% complete a purchase through one. Forrester VP and principal analyst Chuck Gahun calls that 21-point gap the single thing holding agentic commerce back.
The 4GPTs read: the gap is not a demand problem. More than a third of familiar consumers already reach for an agent to shop. What stalls at checkout is trust, and trust is downstream of whether the agent can read, price, and stand behind the product it found.
Why the sessions that do arrive convert so well
The traffic that does cross over behaves differently. In Forrester's own words, answer engines are referring traffic to merchant websites, and that traffic tends to convert very well. The interview that surfaced the report framed it as fewer visitors, roughly three times the conversion, a directional figure rather than a precise multiple.
The reason is composition. A human landing on a product page sits somewhere between idle curiosity and ready to buy, and your conversion rate carries all of it. An answer engine does the filtering first: it matches intent, compares options, checks price and availability, and rules out poor fits. The session that reaches you is already pre-qualified.
Selection is the constraint
That higher-converting cohort only flows to brands an agent can parse. If your best product information lives in prose a human reads, or behind a script an agent never runs, you are invisible at the moment selection happens.
The trust gap is a liability gap
Ask why people discover but do not buy, and the answer is accountability. When an answer-engine purchase goes wrong, 44% of consumers say they would hold the answer engine itself liable, not the merchant and not the payment provider.
That is a market still building its guardrails: returns, disputes, policies, and the structured signals that let an agent trust a seller enough to transact. Reputational signals an agent can verify, not marketing claims a human skims, are what move a brand from discovered to bought.
Forrester's own answer: build a machine advantage
Forrester does not leave merchants guessing. Its playbook has four parts.
1
Agentic fit
Assess where agents can actually serve your catalog.
2
Machine advantage
Target machines with schemas, comprehensive product data, and topical depth.
3
Organizational alignment
Break the silos and build a culture that experiments.
4
Operational revamp
Rework commerce operations for agent-driven demand.
Build a machine advantage to get AI agents to favor your brand, through schemas, comprehensive product information, and at the right level of topical depth.
Read the second item again. Schemas, comprehensive product data, topical depth, context, fluency, and trust. That is a description of an Agentic App: a structured interface a business exposes so agents can find, understand, and recommend it. The point worth sitting with is that Forrester is describing the 4GPTs thesis in its own words, and reputational ranking, being chosen on quality an agent can verify rather than ad spend, is what that machine advantage buys.
The B2B tell: old spec pages are gold mines
The clearest signal of where this goes is which pages already work.
Those spec-heavy, JavaScript-free product pages everyone mocked turn out to be exactly what agents want.
B2B, full of exactly those pages, is accelerating faster than consumer retail. Forrester expects a fifth of B2B to move to human-authorized, agent-executed transactions by year end.
The 4GPTs read: machine-readable beats pretty. The pages that lose in the agent era are the ones tuned for a human eye and a marketing team. The pages that win read like a spec. Forrester's companion research argues the retail website is not dead either, it is becoming a repository of data for agents, and that shift toward storefronts agents can read is already underway.
What to do this quarter
You do not need a moonshot. You need to be legible to machines and to measure the cohort that already matters.
- Expose clean, structured product data. Treat it as a first-class deliverable, not an SEO afterthought: complete attributes, current pricing, real availability, stable identifiers.
- Treat your feed as the storefront. For agent-mediated demand, the feed is the shelf. A thin or stale feed is a closed store.
- Measure agent-attributed conversion separately. Do not let the small, high-converting agent cohort hide inside a declining top-line traffic number.
- Instrument which agents surface you, and which do not. Selection is now a channel; track it like one.
FAQ
Forrester's mid-2026 read finds strong discovery but weak purchase: 38% of consumers familiar with answer engines use them to find products, and 17% buy through them. Hype is running ahead of behavior, and true autonomy, an agent buying without oversight, is still rare.
Stop optimizing for visits, start optimizing for selection
The old game rewarded whoever bought the most attention. The agent era rewards whoever is easiest for a machine to choose. Those are different skills, and the brands that internalize the difference first will compound an advantage while competitors keep chasing a traffic number that no longer means what it used to.
Make your catalog legible to agents
4GPTs makes brands parseable, trustworthy, and selectable by the AI agents doing the choosing. If fewer-but-better sessions are the future you are planning for, the work starts with being readable to the systems doing the selecting.
Sources: Forrester, "The State Of Agentic Commerce In Mid-2026" (Emily Pfeiffer, with Chuck Gahun, Joe Stanhope, and Joe Cicman). Retailgentic, "Fewer Visitors, 3X the Conversion: Chuck Gahun on The State of Agentic Commerce (Q2)". The 38 / 17 and 44% figures are Forrester's, reported via Gahun; the "3x" framing is directional.